Sun, Fujitsu launches entry quad-core Sparc box
Mostly for developers, sometimes for apps
Fujitsu and server partner Sun Microsystems will today roll out an entry server based on the quad-core Sparc64 VII processor created by Fujitsu.
The server, code-named "Ikkaku" (Japanese for "narwhal") and sold as the Sparc Enterprise M3000, is a single-socket box that will fill in a product gap in the companies' combined Sparc T and Sparc64 VII product lines.
The Sparc Enterprise M3000 is a single-socket box, and the four cores in that single Sparc64 VII processor run at 2.52 GHz. The processor has 64 KB of L1 data cache and 64 KB of L1 instruction cache per core, and 5 MB of L2 cache on chip shared by the four cores. The motherboard in the system supports up to 32 GB of main memory using 4 GB memory, and has four low-profile PCI-Express x8 peripheral slots. The system uses the same "Jupiter" server bus as larger Sparc Enterprise M servers to link the components of the system together. That system bus has 17 GB/sec of peak aggregate bandwidth and 4 GB/sec of I/O bandwidth. The server comes in a 2U rack-mounted form factor and has room for two 2.5-inch SAS drives and a DVD drive - the disks come in 146 GB capacities.
According to Sun, the M3000 has about twice the performance of the entry servers using its UltraSparc-IIIi processors from a few years back, but uses about half as much energy and space as that old Sun box.
John Fowler, the executive vice president in charge of Sun's Systems Group, says that the M3000 is being brought to market for a few reasons. For one thing, customers who deploy larger Jupiter systems usually do so with an n-tier architecture, with bigger servers running the databases behind the applications and Web and application servers accessing the data and running the application code that feeds and feeds off the databases. While some companies don't mind mixing different kinds of database and application servers, others do and Sun and Fujitsu needed a smaller server to be an application tier for midrange customers.
Moreover, companies deploying Jupiter machines in the data center want to give developers the same iron and Solaris software stack to work on, and paying for a midrange machine or a slice of a larger system is a lot more expensive than a dedicated entry server. The machine is rated at 47 decibels, which Sun says is at about the same level as a quiet office, so programmers can tuck one into their cubicles. A tower server would be more programmer-friendly, of course. Then again, there is no reason why the programmer needs to sit next to the server he or she is using to develop code - that's what we have networks for.
And still other customers who use Sparc iron want more single-threaded performance in an entry box than the multi-core "Niagara" family of servers deliver. Some applications like threads and don't mind the relatively low clock speed of the Sparc T series of chips, while others do. The M3000 is for those apps that do mind.
There was some talk in Sun and Fujitsu presentations about the M3000 being used in HPC clusters, too, and while Fowler did not deny that this might happen, he didn't sound like it was business that Sun was counting on. Fujitsu's chances for selling clusters of M3000s might be better back in Japan.
The Sparc Enterprise M3000 is available now. It supports the Solaris 10 10/08 Update that is due any day now. A base machine with the single processor, 4 GB of main memory, two 146 GB disks, a DVD drive, and a Solaris 10 license costs $15,000.
With the M3000, Sun and Fujitsu have launched a second generation of their "Advanced Product Line" jointly developed Sparc iron, and that seems to be it until next year, when the Sparc64 VII+ kickers come out. You can read what little has been said about these future chips here. ®
RE: Re: RE: Re: RE: Oh bugger....
"....Why the hell would Sun benchmark dual core Sparc64's when they are selling quad core Sparc64vii that can have 256 cores in a 64 socket frame...." True, but then Sun have to have twice as many cores to beat Itanium, and that means it will be more expensive - much more expensive. Which is one of the reasons why more customers buy Itanium in Integrity servers. Other reasons would be a better support structure, better designs, and management tools that actually integrate and work. You also forgot to mention that when Sun populates FSC's frames to capacity they lose expansion options, which doesn't happen with the HP Integrity range. Which is why the market is beating your beloved Sun down to a delisting.
"....If/when quad core Tukwilla comes out do you not wonder that the Sparc64 might be onto 8 cores...." Well, you should check the rather vague FSC roadmaps, then. FSC don't plan on having the speedbumped Jupiter+ until mid-2009, which means in Sun servers about Q1 2010 at the earliest. There is no date for FSC eight-core parts, they seem have to have caught the vapourware bug from Sun. Which means Intel have until forever to get Tukzilla out the door, but instead they have a solid roadmap showing Tukzilla due Q1 2009. The 32nm process for Tukzilla and the CSI aka Quickpath has already been demonstrated in the latest Xeon CPUs, so the last technical hurdles have been cleared, whereas FSC haven't shown they can shrink their process or make a working four-core. So, in short - again - you just showed your lack of industry knowledge.
"....16core Rock CPU's could be out before then...." LOL! Which roadmap are you looking at, because not one analyst (Gartner, IDC, etc) thinks that. If Sun do release a Rock chip next year it will be a crippled design because they still haven't sorted the technical issues bedevilling it. And given Sun's long history of slippage to the n-th degree, Rock is much more likely to just get canned. Remember UltraSPARC V? Denial is not a river in Egypt, it's running through your head.
"Itanic 2, sunk by Rock instead of Icebergs".... I think you're much more likely to see the headline "Sun sunk by Schwartz". Just take a look at the relative market positions of HP, Intel and Sun, and the share prices for the three, and it is blatantly obvious which one is dying. Better polish up your resume and learn some Linux!
"....Lets do some link exchanging again shall we?...." What, Sun trying to talk about blades!?!?! Don't make me laugh. Sun's blade strategy is as flawed as the rest of their product lines, which is why they have less marketshare than Dell. Oh, by the way, the market leader BECAUSE THE CUSTOMERS RATE THEIR KIT THE BEST is hp, not Sun. IBM sell more blades than Sun, Dell do too (well, now they have copied the hp c-class design). And the chip that powers the most popular blade design is Xeon, not any Niagara. You seriously believe Niagara blades will rule the market? Dream on. They have no chance, which is why Sun have been forced to make Xeon x86 servers. But, seeing as you want a link, I won't go and get an hp blog link or anything that might be construed as vendor biassed, I'll just post the Gratner article here at the Reg (http://www.theregister.co.uk/2008/08/22/gartner_server_q2_2008/ ) - see the tiny little green slice compared to the big blue and red slices? Yup, Sun is the minnow, HP and IBM are the big fish. Read it and weep.
"....Like before, yes, this is a slower frequency than the latest but look at the curves HP boy, 8 threads and BOOM!! Bye bye Itanium performance...." What, Sun trying to benchmark someone else's kit? Yeah, I'm sure that was an unbiased experiment! LOL! How about another real world example, but seeing as you don't like mine I'll borrow one from AMDOCS, the guys that run just about all the telco billing systems in Europe. The AMDOCS ClarifyCRM app was originally written on Slowaris for SPARC. Nowadays it gets implemented on hp-ux on Itanium on Superdome at just about every major telco. The reasons are simple - Itanium scales better than any SPARC, with better performance, with simpler implementation and better management, and with a better return on investment. Telcos used to be Sun's happy hunting ground, now they are losing share to Itanium and Power to the point where a telco deploying any billing app on Sun is an exception. Go read the AMDOCS case studies. Face it, you ask for my love but you really need to give some love back to your customers becasue - as the market figure show - they don't love Sun anymore.
Re: RE: Re: RE: Oh bugger....
"256 cores in the Sun (sorry, I meant FSC) system, compared to 128 cores in the Superdome"
Er, perhaps thats because Itanium has only managed to get to a dual core model. Why the hell would Sun benchmark dual core Sparc64's when they are selling quad core Sparc64vii that can have 256 cores in a 64 socket frame. Itanium can only hold 128 cores in a 64 socket frame, your problem (or HP's it would seem). Sorry it beat your beloved.
If/when quad core Tukwilla comes out do you not wonder that the Sparc64 might be onto 8 cores, hell, 16core Rock CPU's could be out before then, there still scheduled. Quad core Itanic's, 16 core Rock, this could be a call to Hollywood for a new film :
"Itanic 2, sunk by Rock instead of Icebergs". I want the lead role!
Anyway, onwards : "Xeon can already scale to more than Niagara"
Come on Matt, show me the love, I'm begging!!! Lets do some link exchanging again shall we? OK, here goes mine again : http://blogs.sun.com/mrbenchmark/
Like before, yes, this is a slower frequency than the latest but look at the curves HP boy, 8 threads and BOOM!! Bye bye Itanium performance.
Isn't this what led you in the first place to state T2's were faster than Xeons?
Go on Matt, do it again, show me the love....
RE: Re: RE: Oh bugger....
"Comparing core to core is a waste of time...." No, most enterprise software is billed per core, so unless you can build your stack from free software you will pay more to get the same performance with Sun kit as you will with Integrity. And even if you do get a fully free stack, you still have to pay someone to support it and - guess what - most software support costs are per core. So the Sun solution - sorry, I mean the FSC solution in drag - would still cost you more than the Integrity solution.
"....Get an Oracle site license..." Another bit of distortion - there is no such thing as an Oracle site licence, each one is actually a carefully calculated individual licence based on how many users and/or how many cores Oracle expects you to use. Oracle site licences are subject to revue and audit, so simply pretending that there is a one-fee-buys-all fixed licence is not true - if Oracle think you are going to use more sockets they will want to charge you more.
"....Also, as has been pointed out before, what makes you think that Tukwila will ever come out?...." And I give the usual reply - what sustainable argument do you have against it coming out? None. Both HP and Intel make a profit off Itanium so they both have an interest in bringing out Tukzilla. Compare this to Sun, whom are losing money faster than they can make it, and although Sun need Rock you suddenly realise they may not be able to afford it by the time it is ready.
"....My guess is that Itanic is Xeon bound, which will reduce HP's scalability even more...." Well, Bill, your guess just goes to show your lack of technical and industry knowledge. Xeon can already scale to more than Niagara, and using grid technology has scaled to greater size than any SPARC installation in existance. That didn't stop Itanium being produced. What you should consider with real fear is that CSI (aka Quickpath) will allow Xeon to scale to similar reaches as Itanium without needing grid technology, which will just about kill any chance Niagara has, especially with the low-power tech from Atom in the mix. A future low-power Xeon using CSI and HP's cell-based technology could scale to 256 cores, and I bet it will come in a nice blades chassis too. Users will choose Xeon blades with Windows and/or Linux for normal tasks, and Itanium blades with hp-ux and/or Linux for heavy thread tasks, and maybe Atom blades for low-powered tasks like virtual desktops. I'm sure Sun will try and follow behind IBM and HP as they get their offerings out first, but I doubt Sun's hardware division will last that long. By then, Sun will just be a software shop flogging support for Slowaris on Xeon, and no-doubt still losing market share to Red Hat.